-Fixed an issue with the weights argument where, in some cases, custom weight columns were not recognized, causing equal weights to be assigned to all observations. Now, users can specify any column name as a string for weights in datahp and datahs, with the function correctly applying the specified weights for improved flexibility and usability.
This update introduces significant improvements in the flexibility, accuracy, and functionality of the package. Below are the primary modifications:
Correction in handling the dependent variable
(y): An error in processing the dependent variable
led to inaccuracies in estimations. This issue has been resolved,
ensuring reliable and consistent results.
Extended compatibility for the y
variable: The y variable can now be a categorical
variable with multiple levels (J > 2). When
y is a dummy variable, the first column represents level 1.
For categorical variables, levels are ordered alphabetically.
Correction in handling the error prior
(l): Previously, the value l did not
include index 3. This version corrects that omission.
Enhancements in q assignment:
q can now be specified as a vector rather than only a
single value per category. The function now correctly handles uniform
distributions regardless of the number of categories.
New default tolerance level: The function
tolerance is set to 1e-10 for improved optimization
accuracy.
Change in optimization method to
nlminb: This version uses nlminb
instead of optim, improving results and eliminating the
need for the method argument.